Receipt OCR stands for Receipt Optical Character Recognition. It refers to any technology transforming an unstructured image or pdf of a receipt into structured data. This technology can be distributed as a software for developers, on the cloud (API) or as a library, allowing them to build receipt scanning features in their applications and avoid manual data entry.
OCR refers to technologies capable of detecting and reading text from images or documents in order to transform them into machine-readable format. More details in our blog.
When used alone, the acronym OCR traditionally refers to the generic problem of detecting and reading text in an image, independently of any context. When used with additional context such as Receipt OCR, the term doesn't refer to generic text extraction, but to the extraction of key information in a Receipt image. This pattern can be extended to other document types. Examples: Invoice OCR or Passport OCR.
While receipt OCR refers to the action of extracting key information from receipts, receipt scanning is the wider process that includes the capture of the receipt and the information extraction.
Recent receipt OCR technologies combine Computer vision (CV) and Natural Language Processing (NLP) to detect and recognize text inside images. Today, most of those technologies are based on machine learning or deep learning.
Like humans, our algorithms donβt need to read all the document text in its language to extract the relevant information
Our OCR transforms any photo or scan of a receipt into usable data in your software
Total Amount
Total spent including taxes, discounts, fees, tips, and gratuity. The Receipt OCR API supports both typed and handwritten characters for this field.
Total Net
Total amount of the purchase excluding taxes, tips, gratuity, discounts, and other fees.
Tip & Gratuity
Total amount of tip and gratuity. Both typed and handwritten characters are supported.
Taxes breakdown
Date and time
Receipt's payment timestamp:
Category
Purchase category among a list of 7 possible values: food, toll, hotel, gas, transport, parking, other
Supplier Information
Currency and Locale
Finding the right OCR technology to use for your project can be a heavy task. Whatever your use case is, criteria like extraction performances, response time, integration time, pricing, scalability... should be taken into account in order to maximize the added value in your software. Feel free to contact us if you don't find the answers to your questions below.
Our receipt API is free to use and available to any user having an account on our platform.
To test our APIs, you only have to create a free account using this link, and you'll be able to drag and drop receipts in the live interface to see the data extracted in real-time and JSON response. A demo page is also available here.
A free plan is available to everyone and allows you to perform 250 receipts processing per month for free. No credit card is required.
Above 250 receipts per month, the price per receipt processed starts at $0.10 and can decrease to $0.01 per receipt depending on the monthly volume. See the pricing page for more information.
Our receipt OCR API is based on our computer vision technology that doesn't rely on text to extract the receipt data, but only on the image. This removes the language limitations.
The OCR was trained with receipts from more than 50 countries and works on receipts from all around the world for numeric fields, and all Latin alphabet countries for text ones.
Mindee's API follows HTTP standards in order to allow any developer to integrate the receipt OCR API into their applications easily.
We also offer a set of client libraries in all the main back-end languages, and an open-source UI toolkit that helps create front-end features. You can check out our open-source repository or our API documentation for more details.
Our receipt OCR's accuracy is above 90%, with precision above 95% for most of the fields. These performances are computed on a data set including more than 50 countries.
Testing our OCR API is free, all you need is an account. Feel free to drop receipts in the live interface to see the OCR performance on your data.
The processing time is around 1.3 seconds per page for pdfs and 0.9 seconds for a receipt image.
We often improve this processing time and our target is below 500ms. Our goal is to make sure you can create real-time user experiences in your application.
Yes, the OCR was trained on a lot of receipts from a wide variety of layouts and image quality and learned to process the most complex ones.
We also use data augmentation to make sure that no blur or ink stains prevent the OCR from reading the data as long as it's readable.
We have a Slack community where you can ask your questions and chat with our team.
We don't do the integration in your infrastructure ourselves but e can set up a custom level of support on a per-case basis if needed.